# Research Papers — 2026-09 · bundle page 69 of 266 > Full-text markdown of papers #3401–3450 of 13298 published in 2026-09, newest first. AI-readable mirrors of the full 13298-paper 2026-09 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-09/page/69 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/68.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/70.md) ### LLMAdBench: A Human Preference Benchmark for Advertising in LLM Responses Authors: Rui Ai, Yuqing Liu, Sitao Qiu, Yun Qiao, Yuhan Wang, Jessica Xiwen Wang, Yiqi Yang, Lihong Huang, Ruiyao Sun, Kaifeng Zhang, Shengze Ding, Jiaqi He et al. Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32533v1 Inserting advertisements (ads) into consumer-facing LLM output is emerging as a new business model, but there is little shared evidence on how such ad insertion should be evaluated or how it affects user preferences. We introduce LLMAdBench, a human-preference benchmark for studying advertising in LLM-generated content. The benchmark isolates a simple but practically important decision: given a user conversation, an LLM response, and a matched advertisement, where should the ad be placed? Our dataset compares pairs of responses that differ only in ad position while holding all other conditions fixed including the user query, base answer, advertisement, and disclosure condition. Human annotators evaluate each pair based on six criteria from both advertiser's and user's perspectives. The resulting benchmark contains more than 18000 human judgments across two disclosure conditions: explicitly labeling the ad as sponsored and merging it into the response without disclosure. We use LLMAdBench to evaluate eight frontier LLMs as preference judges and find that they are not reliable substitutes for human evaluation. Even the most stable models reverse roughly one quarter of their decisions when the presentation order is swapped, agreement across models is low, and their placement preferences differ systematically from those of human annotators. Moreover, LLMAdBench contains substantial learnable signal. In particular, a Qwen3-8B model fine-tuned on the human preferences improves substantially over its base model and outperforms all zero-shot frontier judges on the held-out prediction task. Beyond model evaluation, LLMAdBench provides quantitative evidence on the advertiser-user trade-off and shows that the sponsorship disclosure systematically changes users' preference over ad placement. --- ### What Does a ProcGen Generalization Gap Measure? Action Rules, Convergence, and the Missing Random Floor Authors: Abhisek Keshari Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32532v1 A generalization gap in reinforcement learning (return on training levels minus return on held-out levels) is usually reported without a reference point. We argue that the missing reference is a measured random floor: the return of a uniform-random policy on the same levels under the same harness. On ProcGen, the floor changes what several standard numbers mean. On identical checkpoints and levels across eight environments, switching between sampled and greedy (argmax) test-time actions moves held-out return in both directions, and greedy evaluation takes three environments to or below the floor: in miner, the sampled policy scores 4.9x the floor on held-out levels while its argmax scores below it. Raw policy entropy places seven of eight environments short of convergence, but 35-65% of that entropy is spread across actions with identical effects; after merging them, one to three remain short, and against the floor only heist has learned nothing that transfers. An audit of twelve prior ProcGen codebases finds that all eleven with held-out evaluation sample test-time actions for their policy-gradient agents, nine by default rather than explicit choice, and six report running in-loop averages rather than evaluating a fixed checkpoint. Applied to our own case study, the same checks grade down a statistically significant encoder effect and rule out a within-encoder train-vs-test CKA statistic. We recommend that every reported gap state its action rule, use a matched and seeded protocol, and report the random floor on both level sets. --- ### Activation Flow: Manufacturing Activations for Steering Authors: Hong Kiat Tan, Linh Le, David Williams-King Published: 2026-09-26 Categories: cs.LG, cs.AI, cs.CL, math.DG arXiv: https://arxiv.org/abs/2609.32530v1 Difference-in-means steering requires activations recorded while a model shows the desired behavior, which a sandbagging model withholds by deliberately underperforming. We introduce Activation Flow (ActFlow), which manufactures these activations from $k$ correct labels without fine-tuning. ActFlow sets target logits that rank each labeled item's correct answer first, and moves the logits toward them by adding one vector $x$ to all $k$ residual streams at one layer. ActFlow is a family of ordinary differential equations for $x$, one for each rule that maps the required logit change to the velocity of $x$. The smallest-norm rule lands exactly on the targets, while the others keep only the top singular directions of the Jacobian. We test ActFlow on three instruction-tuned models, each locked by a sandbagging prompt and by a password-locked LoRA. At $k=40$, ActFlow keeping five singular directions raises the mean held-out ARC-Easy accuracy over the six locked models from $0.05$ to $0.85$, against $0.88$ for fine-tuning and $0.92$ for the honest models. Furthermore, it scores higher than the smallest-norm rule in 16 of the 18 combinations of locked model and $k$, and its steering direction is nearly orthogonal to the honest difference-in-means direction. It also unlocks two LoRA locks where the honest direction fails. --- ### SetOPD: From Few Visual Exemplars to Multimodal Candidate Sets for Remote-Sensing Open-Prompt Detection Authors: Jinlong Hu, Yi Zhang, Zhiqi Xia, Yikang Zhou, Shunping Ji Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32529v1 Open-prompt detectors allow users to specify targets with text, visual exemplars, or both. We argue that existing designs underuse the visual modality in two ways. First, multiple exemplars are commonly compressed into a single class-level embedding. This textualizes visual prompting: the resulting vector plays the role of another class name, is often aligned to or injected into the text pathway, and may be suboptimal when only a few heterogeneous exemplars are available. Second, existing methods interact primarily in prompt or representation space, before modality-specific detection states are formed. We address both issues from a set perspective: \setopd preserves modality-specific decoding states from a shared prompt-conditioned initialization and performs explicit multimodal collaboration at the candidate-state level. For the first issue, we introduce \br prompting, which reads every boxed exemplar in its full scene context and decomposes the pooled support evidence into a Base anchor and a learned Residual correction; the resulting prompt has fixed capacity regardless of the number of examples and drives its own visual detection pathway. For the second, we recast text--visual collaboration from representation-level fusion into a candidate-set modeling problem. Paired-Query Arbitration (\pqa) then performs explicit cross-modal state arbitration only after modality-specific candidate states have been formed. The two readers share query initialization so their candidates are paired by index; a learned gate arbitrates within each pair, followed by a permutation-equivariant module that reasons over the fused set. --- ### Fail Loudly: An Auditable Runtime for Agentic Data Analysis Authors: Hanxu Yan, Langxuan Deng, Zhengle Wang, Yibo Wang, Chunwei Liu Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32528v1 Large language models (LLMs) have enabled data-science agents to automate multi-step analyses over heterogeneous files. However, incorrect choices regarding data sources, scope, or statistical definitions often lead to silent errors: computations execute successfully but produce plausible yet incorrect outputs that fail to answer the intended question. To mitigate this, we present RADAR, an auditable runtime that makes an agent's analytical choices inspectable and supports their revision through execution feedback. RADAR operates through three core mechanisms. First, an evidence-preserving exploration module retrieves task-relevant content while retaining source locations and observation coverage. Next, the runtime uses typed operators to record the agent's declared inputs, operation arguments, and resulting observations. Finally, runtime validation checks proposed operations against these observations. When a conflict is detected, the runtime rejects the operation or provides diagnostic feedback, allowing the agent to revise its choices before errors propagate. This design enables agents to fail loudly while leaving semantic interpretation to the LLM. On KramaBench, RADAR achieves overall scores of 0.723 with full source retrieval and 0.747 with gold sources supplied, corresponding to relative gains of 35.9% and 28.8% over the strongest baselines. Beyond KramaBench, RADAR achieves relative performance gains of 14.0% on DA-Code and 59.3% on DABStep, demonstrating its applicability across diverse agentic data-analysis workflows. --- ### AmbiModBench: Benchmarking Gene Perturbation Prediction Beyond Shared Responses Authors: Sikai Huang, Zhiwen Yang, Kai Yu, Jiayuan Chen, Stan Z. Li Published: 2026-09-26 Categories: cs.AI, cs.LG, q-bio.GN, stat.AP arXiv: https://arxiv.org/abs/2609.32527v1 Predicting cellular responses to genetic perturbations helps prioritize experiments in single-cell genomics, where exhaustive measurement is infeasible. While computational models increasingly predict these responses, three evaluation deficiencies obscure what their scores demonstrate. First, absolute metrics cannot separate target-specific predictions from a shared background response. Second, common metrics remain high under gene shuffling, so gene-level accuracy is never verified. Third, a score at one training size says nothing about coverage, which depends on representation-space proximity and response-constraining power. We propose AmbiModBench, a specificity-aware, gene-resolved and coverage-aware benchmark. It pairs every score with a training-mean reference fitted on the same split, screens each readout by gene-coordinate permutation, and links embedding distance to response variation. Across K562, RPE1 and Norman, strong absolute scores largely reflect shared background rather than target-specific learning. Widely used readouts track response magnitude distributions rather than the affected genes. Detectable gain follows representation-space coverage rather than training-set size. Nonetheless, on RPE1 the protocol yields a reproducible target-specific gain across five additional splits and three gene selections, which absolute scores alone cannot distinguish from shared background. --- ### Does Transolver really need a Transformer? Authors: Shizheng Wen, Siddhartha Mishra Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32525v1 The widely used Transolver family of neural operators is based on physics-attention, which softly assigns the points of an unstructured mesh to a small number of slices, applies self-attention among the resulting tokens, and broadcasts the result back to the points. We provide a comprehensive empirical and theoretical analysis to elucidate the mechanisms which are responsible for model performance. To this end, we perform careful ablations on a challenging suite of nine 3D fluid dynamics benchmarks to find that replacing token attention with a constant linear map does not affect the accuracy. Thus, Transolver does not need a Transformer at all. However, removing the global mixing (slicing/deslicing) or doing it only once leads to performance collapse. We leverage the theory of averaging neural operators to explain and corroborate our findings by showing that just slicing/deslicing, in conjunction with pointwise MLPs, already suffices for universal approximation of continuous operators and attention is redundant in this context. Finally, we provide a novel FlashAttention-style efficient implementation of the key slicing/deslicing module of Transolver. This flashslice kernel streams slice and deslice over the points without materializing the heavy slice-weight tensor, while reproducing the best available implementation to floating point error. At the same time, it leads to very significant memory and compute savings, particularly at large slice counts. --- ### Beyond Dyadic Memory: Interaction-Aware Multimodal Memory with Adaptive Agentic Retrieval for Multi-Party Spoken Conversations Authors: Wenxu Jia, Xize Cheng, Zihan Zhang, Dongjie Fu, Linjun Li, Wenshi Chen, Yangyang Wu, Tao Jin Published: 2026-09-26 Categories: cs.AI, cs.IR, cs.SD, eess.AS arXiv: https://arxiv.org/abs/2609.32522v1 Long-term memory enables agents to accumulate information and reason across sessions, yet existing research primarily focuses on dyadic text or image-text conversations, leaving long-term memory for multi-party spoken conversations underexplored. This setting requires preserving conversational content, identifying participants across sessions, and retaining who speaks to whom. To this end, we propose VoxPolyMem, an interaction-aware multimodal memory framework combining incremental speaker identification with a memory hierarchy comprising interaction memory, fact memory, and participant profiles. We formulate retrieval as sequential decision-making, where an agent rewrites queries and selects retrieval tools and memory layers based on accumulated evidence to address information gaps. We further introduce Evidence-Gain GRPO (EG-GRPO), which uses round-wise credit assignment to encourage complementary evidence acquisition. We also construct VoxPolyBench to evaluate memory evolution, personalized answering, memory retrieval and reasoning, and interaction reasoning and attribution in multi-party spoken conversations. VoxPolyMem achieves an overall score of 85.0 on VoxPolyBench, surpassing the strongest evaluated baseline by 23.6 points. On Mem-Gallery and H2HMem-Multi, it scores 89.6 and 74.4, respectively, exceeding the strongest evaluated public memory baselines by over 8 points each. These results highlight its potential for persistent, personalized assistance in multi-party multimodal interactions. Code and datasets are available at https://voxpolymem.github.io/VoxPolyBench/demo/ --- ### MemAgent: Learning to Manage Heterogeneous Memory Providers for LLM Agents Authors: Yongxian Wei, Yilin Zhao, Runxi Cheng, Xinrui Chen, Chun Yuan, Yaoru Wang, Jiahong Yan, Dian Li Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32521v1 Current agents remain largely stateless across tasks, limiting their ability to continually improve from prior interactions and making memory essential for long-horizon agentic behavior. Existing memory methods seek to reuse past experience, but most rely on a single memory representation (e.g., trajectories, reflections, skills, structured knowledge) whose effectiveness varies across task distributions. Rethinking this design space, we evaluate 13 memory methods and find that no single method generalizes across benchmarks, revealing the potential of managing heterogeneous memory providers. We formulate agent memory as a routing problem in which a memory agent decides which memory provider to retrieve from, whether to inject short-term memory, and which providers should store the resulting experience. Based on this perspective, we propose MemAgent, featuring a content-aware routing architecture and a training-data synthesis pipeline. The routing architecture combines content-aware probing before retrieval, short-term memory gating during execution, and selective multi-provider storage, while the training pipeline synthesizes phase-specific supervision for routing decisions. Across GAIA, WebWalkerQA, and xBench-DS, MemAgent improves average accuracy by 10.0% and outperforms every individual memory method across all three benchmarks. These gains come with less than 0.3% routing overhead and a 12% reduction in average task steps. --- ### When Users Change Their Minds: Measuring and Repairing Intent Drift in LLM Agents Authors: Yanjie Zhang, Bowen Cao, Zixin Chen, Yushi Sun Published: 2026-09-26 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.32520v1 LLM agents often operate over multi-turn interactions in which user intent changes before execution. We study intent drift: the failure mode in which superseded parts of the user's intent continue to influence the final answer or tool action. We introduce IntentFlux, an executable benchmark that converts verifiable tasks into dialogues with controlled intent changes while preserving their original graders. In a 627-case calibration, mean task score falls from 0.476 to 0.384 as dialogues contain more superseded and withdrawn information. Across eight models, the rate of fully correct solutions is significantly lower when the same final task must be recovered from an evolving dialogue rather than given directly in a single turn. We further introduce StateForge, which explicitly maintains the active requirements before generation. On General-Test, it improves mean task score from 0.367 to 0.467. Providing the ground-truth final state improves performance further but still does not recover single-turn performance, indicating that state-estimation errors explain only part of the gap. These results establish intent drift as a measurable multi-turn failure mode and explicit state maintenance as a partial mitigation. --- ### STR: Supervised Transcoder Replacement for Reducing Steering Side Effects Authors: Haonan Yu, Junhao Liu, Zhenyu Yan, Haoran Lin, Xin Zhang Published: 2026-09-26 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32519v1 Model steering can strengthen a target behavior while degrading other useful behaviors. We introduce Supervised Transcoder Replacement (STR) to reduce these side effects for existing steering methods, including those fitted without a protection objective. STR learns a replacement for the multilayer perceptron (MLP) computation at the steering layer through supervision for target control, non-target preservation, and fidelity without steering. Selected steering methods then fit directions on the frozen replacement while retaining their own fitting objectives. We evaluate three steering methods across Gemma and Llama models using Corrigibility preferences and four harmful-request safety datasets. SALAD-Bench supplies protection training data and a separate in-distribution evaluation split; HarmBench, AdvBench, and StrongREJECT are reserved for out-of-distribution testing. STR substantially reduces steering side effects on the in-distribution evaluation and extends this protection to the unseen safety datasets while retaining effective target control. For target-only supervised steering vectors, pooled out-of-distribution attack success rate falls from 42.46% to 14.42% on Gemma-3-4B and from 34.97% to 12.91% on Gemma-3-12B. These results show that replacement training can benefit steering methods fitted without protection objectives. --- ### UnStep: Training-Free Acceleration of Causal Video Diffusion with Fewer Steps Than Distillation Authors: Youssef Mansour, Enis Simsar, Fadime Sener, Markos Georgopoulos, Albert Pumarola, Ali Thabet, Edgar Schoenfeld Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32518v1 Distilling bidirectional multi-step video diffusion transformers into few-step causal models has become a common approach for streaming video generation. While these few-step students are significantly faster than the teachers they are distilled from, they remain slow for real-time generation. In this work we present UnStep, a training-free wrapper that accelerates few-step causal video models at inference by running them with fewer diffusion transformer (DiT) steps than during distillation and limiting the temporal window retained in the attention KV cache. We propose two inference-only mechanisms to recover quality lost by step reduction and attention windowing: renoising the generated latent frames to a near-clean level and reusing the existing clean-cache pass to refine them, and applying truncated SVD to the DiT attention value and output projections. We also accelerate inference with a quality-preserving runtime stack for the DiT and VAE decoder, including more efficient attention calls and KV indexing, fused Triton RoPE with cached coefficients, and VAE decoding with optimized memory layout, precision, and convolution kernels. By reducing computation and optimizing the runtime stack, UnStep sets a new throughput regime for causal video diffusion, by running substantially faster than current methods, reaching 50 FPS on a single H100 without quality loss, and 77 FPS on GB200, all without retraining. --- ### LocalProp: Neuro-Localized Memory-Efficient Backpropagation Authors: Diana-Nicoleta Grigore, Iuliana Georgescu, Radu Tudor Ionescu Published: 2026-09-26 Categories: cs.AI, cs.CV, cs.LG, math.OC arXiv: https://arxiv.org/abs/2609.32517v1 The current deep learning training paradigm employs end-to-end backpropagation, regardless of the training stage, i.e. pre-training or fine-tuning. However, backpropagating through the entire model is neither biologically plausible nor memory efficient, since learning inside the brain is highly localized. Therefore, we propose LocalProp, a training procedure that locally updates the weights of a model. Our neuro-localized weight updates follow the "pre-training then fine-tuning" paradigm, where the pre-training is based on I-JEPA. After locally updating the weights, a pruning operation is performed, followed by a short final fine-tuning phase. Pruning helps by sending the learning signal from higher blocks to lower blocks. We perform experiments on several datasets, including large-scale benchmarks such as ImageNet, and empirically show that LocalProp reaches good performance at a fraction of GPU peak memory. By varying the number of jointly optimized blocks, we identify gradient-propagation span as a practical control over the accuracy-memory trade-off. --- ### REFINE: A Resilient Evolution Framework for Intelligent Enterprise Alert Triage in Security Operations Centers Authors: Huimin Chen, Quan Long, Yanhao Wang Published: 2026-09-26 Categories: cs.CR, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32516v1 Security Operations Centers (SOCs) process large volumes of alerts daily. Alert triage prioritizes high-risk threats while reducing manual review of benign alerts. LLM agents can reason over logs and threat intelligence, but struggle to keep aligned with organization-specific, rapidly evolving SOC operational standards. We introduce REFINE, an LLM-agent framework for enterprise alert triage. REFINE encodes analyst expertise as structured skills and continuously adapts using analyst disposition feedback. It enforces recall = 1.0 as a hard constraint during evolution to maximize auto-closure of false positives, and identifies judgment blind spots by combining alert distributions with model error boundaries. Evaluated on four real industrial SOC scenarios across four MITRE ATT&CK phases with temporal split: REFINE achieves recall=1.0 on all evolution sets. On future test windows, it retains recall=1.0 in three scenarios; the degraded case reaches 0.807 recall, still outperforming self-evolution baselines (0.49-0.58). --- ### From Anomalies to Failures: Constructing Causal Error Graphs for Agentic Trace Diagnosis Authors: Shu-Xun Yang, Yidong Wang, Zhuoer Feng, Bosi Wen, Jiayi Gui, Dayong Yang, Wenbo Yu, Haoke Zhang, Jie Tang, Cunxiang Wang Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32514v1 LLM-driven agents are increasingly deployed in complex applications, where long agentic traces make failures difficult to diagnose. Existing trace diagnosis methods often conflate anomalies, errors, and failures, making diagnostic targets ambiguous; they also lack structured modeling of how causally relevant errors propagate and amplify into final task failures, resulting in unreliable failure attribution. To address these problems, we propose CEG-Agent, a tool-augmented agentic framework for causal diagnosis of agentic traces. Specifically, CEG-Agent introduces an explicit taxonomy of anomalies, errors, and failures, and constructs Causal Error Graphs (CEGs), a unified typed representation that links execution events, diagnostic nodes, and failure outcomes through causal relations. To evaluate causal trace diagnosis, we further construct CEG-Bench, a fully agent-annotated benchmark with high-confidence, consensus-derived CEG annotations obtained through an Adversarial Agentic Adjudication Protocol (AAAP). We validate the resulting annotations against an expert-curated human gold set, which shows close agreement with the automatic annotations. Experiments on CEG-Bench demonstrate that CEG-Agent achieves state-of-the-art performance under both semantically relaxed and structurally exact evaluation criteria. Our code is publicly available. --- ### Cross-Domain Few-Shot Writer Adaptation for Real-World Handwritten Mathematical Expression Recognition Authors: Paulo Grane Gabriel Silva, Lorenz Bernard Marqueses, Joel Ilao Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32513v1 Handwritten mathematical expression recognition (HMER) refers to the task of recognizing and converting handwritten mathematics into a parsable markup language, usually LaTeX. No current state-of-the-art-competitive system adjusts to the way a specific person writes, and the domain gap between training images (usually digital or perfectly binarized) and images physically taken with a camera used in inference has received fairly little attention for this specific problem. We characterize this domain gap through fragmentation and stroke-width analyses of the images as well as introduce a writer-adaptive fine-tuning pipeline to MFH-CoMER in an attempt to address it. We further introduce sample author-specific datasets, consisting of five handwriting category subsets from two authors, and evaluate using a McNemar's test and permutation tests adapted to limited data. Results suggest an increase in expression recognition rate and a decrease in CER for digital handwriting but more varied results for physical handwriting, with the model struggling for handwriting articles that the base model can already evaluate well. Nonetheless, adapted models were found to have improved results for four out of five subsets. Statistical testing results suggest a consistency in improvement for two of the tested author-specific subsets. Our results point towards the potential feasibility of writer adaptation for the HMER task. --- ### What Do Latent Predictive Vehicle Representations Retain? Measuring State, Geometry, and Local Response Authors: Enzo Nicolás Spotorno, Josafat Leal Filho, Antônio Augusto Fröhlich Published: 2026-09-26 Categories: cs.LG, eess.SY arXiv: https://arxiv.org/abs/2609.32512v1 Models of vehicle dynamics learned from logged states and commands complement physics-based models, and latent world models, which predict in a learned representation, are used to plan and train controllers in other domains. Vehicle controllers are usually specified in physical terms: costs, limits, and references depend on position, yaw angle, speed, and yaw rate, and the optimizer compares or differentiates predicted outcomes across nearby commands. A latent model placed in such a controller must therefore let these quantities be recovered and must change its predictions with commands as the vehicle does, and prediction error on its own latent targets measures neither. We contribute a measurement protocol for action-conditioned latent predictors with a physical readout that separately tests retention, physical-neighborhood organization, forecasting, and local response to command perturbations, using an untrained-encoder reference and three matched response paths that locate errors in the representation or the predictor. In a case study of a temporal joint-embedding predictive model trained on signals logged in IPG CarMaker, the representations retain the measured planar outputs, though an untrained encoder of the same architecture retains them slightly better; future-command input improves one-second forecasts with retention nearly unchanged; and responses to small command pulses diverge from the simulator already in latent coordinates, raising regret when choosing among nearby commands in all comparisons. Updating the predictor on responses corrects them locally at a cost in forecast accuracy. Measuring retention, forecasting, and local response separately is thus what qualifies a predictive latent as a candidate model for control, and the protocol provides the basis for its closed-loop evaluation. --- ### Learning from Others, Acting for You: Cross-User Memory Sharing for LLM Agents Authors: Jinming Hu, Haodong Zhao, Qi Jia, Die Chen, Tianhang Zhao, Sufeng Duan, Gongshen Liu Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32511v1 Large language model (LLM) agents serving different users often solve related tasks, yet separate user histories can leave reusable experience inaccessible to other agents. Pooling memories expands access but risks transferring preferences that conflict with the receiving user's requirements. We introduce ShareMem, a memory architecture that shares reusable experience while grounding its application in the receiving user's own preferences. Shared experiences indicate how to act and which preferences to consult; the receiving user's memory supplies their concrete values. Two-stage consolidation refines experience locally before integrating accepted edits into a shared pool. During execution, scope-first retrieval jointly selects local and shared experiences under a common entry budget, while a user-bound channel supports initial and agent-initiated preference retrieval. We evaluate ShareMem across web navigation (Mind2Web), online personalized interaction (VitaBench~2.0), and multi-session coding (MemoryCode) with four backbone models. It improves step success, average task success, and dialogue-macro coding scores, respectively, over matched user-local memory across all four models. Ablations favor two-stage consolidation for smaller shared pools, lower induction token usage, and better downstream performance, and support complementarity between experience guidance and active preference retrieval. Further analyses show that sharing helps most when relevant local experience is scarce, while source quality and cross-user preference interference limit useful transfer. --- ### GeoCR: Learning a Generalist Cloud Removal Prior from Heterogeneous Observations Authors: Jeonghyeok Do, Munchurl Kim Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32510v1 Cloud removal methods are typically specialized to individual datasets and input configurations, limiting reuse across sensors, spectral bands, and observation settings. We introduce GeoCR, a generalist model that unifies RGB-only-based CR and multispectral-based CR from single- or multi-temporal cloudy observations, with optional SAR guidance, within a single network. To accommodate different spectral and sensing domains, compact input and output stems extend a pretrained RGB autoencoder while keeping its encoder and decoder trunks frozen. This shared latent interface enables a single flow transformer to jointly model clean RGB and non-RGB latents, conditioned on separate cloudy-observation streams and optional SAR tokens. Through joint pretraining on the training splits of ten datasets comprising 883,331 cloud-free target images, GeoCR learns a shared cloud removal prior across these heterogeneous configurations. The same pretrained checkpoint supports direct inference without dataset-specific fine-tuning and efficient adaptation through low-rank adaptation (LoRA). We evaluate GeoCR against general image restoration and cloud removal methods on test splits of the contributing datasets under full-band and RGB-only settings. GeoCR achieves the best FID and DISTS on full-band SEN12MS-CR and Sen2_MTC_New and RGB-only CUHK-CR2, outperforming existing models and demonstrating the effectiveness of a reusable generative model across diverse settings. --- ### Fast Differentiable SVD on GPU via Polar Decomposition Authors: Uliana Parkina, Askar Tsyganov, Sergei Kudriashov, Sergey Samsonov, Maxim Rakhuba Published: 2026-09-26 Categories: math.NA, cs.LG arXiv: https://arxiv.org/abs/2609.32505v1 We present a fully GPU-oriented SVD pipeline based on polar decomposition, motivated by iterative methods that rely solely on matrix multiplications, such as the Newton-Schulz iteration. We show that this approach enables up to a $2\times$ speedup compared to standard implementations. Furthermore, we derive a numerically stable backward pass for the polar decomposition and leverage it to obtain a fully differentiable SVD. Our methods are released as open-source implementations in both PyTorch and JAX: https://github.com/fallnlove/cans_svd. --- ### Fisher Simplicity in Kolmogorov-Arnold Networks and Multilayer Perceptrons Authors: Ami Tavory, Meir Feder Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32503v1 Kolmogorov-Arnold Networks (KANs) are motivated in part by interpretability: their learned edge functions can be inspected, pruned, and reduced to symbolic structure. In a fixed-basis KAN, this makes a small or zero basis coefficient look like a certificate of simplicity, much as a dead rectified linear unit (ReLU) marks unused computation in a multilayer perceptron (MLP). Fisher nullity gives a precise statistical notion: a parameter direction is Fisher-simple exactly when perturbing it is invisible under the task distribution. We study when these architectural and Fisher notions agree. For a dead ReLU unit, they agree: the closed activation region makes the associated score directions vanish. For a fixed-basis KAN, they do not. In the single-layer Gaussian case, the coefficient Fisher matrix is a basis Gram matrix under the input distribution and is independent of the fitted coefficients. In a multilayer KAN, Fisher simplicity is graph-path based: the data must reach a basis atom and its perturbation must propagate through the downstream network. We encode these two conditions in an effective edge measure and, under local dictionary independence and effective-measure nondegeneracy, show that zero effective exposure exactly identifies Fisher-null directions within an edge. Controlled diagnostics confirm that zero coefficients can preserve rank while effective path disconnections remove the predicted directions. Coefficient magnitude alone is therefore not a Fisher-based pruning criterion for KANs. --- ### TreeRef-BFN: Equivariance-Free De Novo Molecule Generation based on 2D Topology and Internal 3D Geometry Authors: Ruiqing Sun, Sen Yang, Dawei Feng, Bo Ding, Yijie Wang, Huaimin Wang Published: 2026-09-26 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32502v1 De novo 3D molecular generation jointly models molecular size, topology, and geometry. Most methods pre-sample molecular size and generate Cartesian coordinates, limiting variable-size conditional tasks such as fragment completion and scaffold decoration while often relying on equivariant architectures. Internal-coordinate methods avoid rigid-body redundancy but typically require a known molecular graph or autoregressive construction, which may accumulate errors. We propose TreeRef, a tree-based molecular representation that assigns molecular topology and topology-dependent local 3D geometry to a naturally variable-size tree. RingRef nodes encode ring closures while preserving the tree structure, while Null nodes allow molecular size to emerge directly from node occupancy. Based on TreeRef, we develop TreeRef-BFN, a Bayesian Flow Network with a standard Transformer backbone that globally couples these locally defined variables and jointly generates discrete molecular variables and continuous local geometry. A single pretrained TreeRef-BFN supports unconditional generation and variable-size structure-conditioned 3D generation through masking alone, without retraining. Empirical studies demonstrate strong chemical validity, molecular stability, and diversity, accurate local geometric distributions, fast sampling, and competitive property-conditioned generation, establishing TreeRef-BFN as an efficient and flexible framework for 3D molecular generation. --- ### Rondo: Unsupervised Discovery of Recurring Temporal Structure Authors: Yingtian Shi, Ankith Chandra, Thomas Plötz Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32500v1 Many real-world time series data exhibit structural properties at multiple scales, from short, recurring units to complex sequences composed of these units. Unsupervised discovery of both these components and structure enables the design of intelligent systems that help interpret temporal data, thereby limiting the amount of costly human annotations required. Existing modeling approaches typically overlook the hierarchical structure inherent to many time series, treating recurring patterns at different temporal scales as independent structures. Moreover, most assume access to the complete data sequence and treat discovery as a static process, limiting their ability to evolve as new observations arrive. We introduce Rondo, an unsupervised approach for modeling recurring hierarchical structure in continuous temporal streams. By explicitly constructing vocabularies of reusable units and their recurring compositions, Rondo captures structure shared across complex temporal patterns while refining and expanding its discoveries as the stream evolves. Evaluations on temporal sequences spanning diverse domains and data modalities show that Rondo outperforms existing unsupervised recurrence-discovery baselines, with particularly pronounced advantages in limited-data and continual-stream settings. These capabilities provide a stronger foundation for recurring-pattern discovery, scalable behavior understanding, and adaptive intelligent systems operating on long, unlabeled temporal streams. --- ### Learning an Anchored Prompt Space for Continual Adaptation of Large Language Models Authors: Rongguang Ye, Zhan Zhuang, Yichen Wu, Ming Tang, Kede Ma Published: 2026-09-26 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.32499v1 Continually adapting large language models requires acquiring new knowledge while preserving previously learned capabilities. Jointly adapting model parameters and task-specific soft prompts offers a promising solution, but faces two key limitations: historical prompts may become less effective as the model evolves, while their transferable cross-task relationships are not explicitly learned. We propose Learning an Anchored Prompt Space (LAPS), which preserves historical prompt effectiveness and learns relationships among task-specific soft prompts to facilitate positive transfer. LAPS first aligns historical prompts with the updated model through self-distillation. LAPS then constructs an anchored prompt space whose vertices correspond to learned task-specific soft prompts and whose intermediate geometry is shaped by learnable Bézier control prompts. Once this anchored prompt space is learned, LAPS identifies the best-performing prompt for each task on its validation set, allowing the optimized prompt to draw on knowledge acquired from observed tasks. Experiments on the TRACE benchmark across three Qwen3 model scales show that LAPS consistently outperforms distillation-based, prompt-based, and joint prompt--parameter adaptation baselines, improving average performance while reducing forgetting. --- ### DAAF: From Failure Localization to Editable System Assets in LLM Agents Authors: Xiaoyang Yuan, Qi Liu, Yubin Ruan, Xinyi Mou, Zhuomeng Zhang, Wenjin Wang, Hanying Jiao, Di Wu, Mingye Xu, Yi Bin, Ke Feng, Zixun Sun Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32498v1 Deployed LLM agents increasingly rely on persistent, versioned system assets such as routing rules, knowledge segments, prompt instructions, and reusable skills. Failure-localization methods can identify where an error manifests in an agent or execution trace, but repair requires a different decision: which editable system asset should be changed, and is that change expected to improve the task outcome? We study this gap through component-attribute failure attribution, where diagnosis targets versioned, addressable items rather than execution locations. We propose the Detection-Aware Attribution Framework (DAAF), which learns the effects of valid attribute replacements and amortizes this intervention evidence into deployment-time diagnosis. DAAF combines sparse and noisy failure signals to decide whether intervention is warranted, learns component-type-conditioned replacement effects from controlled replays evaluated by executable task outcomes, and shares supervision across requests with compatible intervention responses. At diagnosis time, DAAF uses only the observed execution, registered candidates, and available failure signals; it requires neither counterfactual replay nor task reward and returns no_change, a repair target, or an unresolved decision when evidence is insufficient. On held-out tau^2-bench Telecom tasks, DAAF achieves 80.72% attribute Hit@1, recovers 62.65% of failed executions while limiting clean-task regression to 3.23%, and reaches 71.93% overall task success. These results show that intervention-grounded attribute attribution can connect failure localization to executable system repair. --- ### Locally Sound, Globally Insufficient: The Local-Global Gap in Multi-Hop Reasoning Authors: Bohao Chu, Hendrik Damm, Qianli Wang, Hui Wang, Shuning Zhang, Christoph M. Friedrich, Norbert Fuhr Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32496v1 Reliable multi-hop reasoning requires more than locally supported steps: a trace can be sound at every reasoning step yet still fail to answer the question as a whole. We call this failure regime the local-global gap (LGG), in which the trace is locally sound yet globally insufficient. Local soundness requires each step to be supported by the available evidence and preceding steps, whereas global sufficiency requires the reasoning trace to align with the question and establish the submitted answer. In a human-adjudicated diagnostic of 2,598 responses across three multi-hop QA benchmarks and three models, we find that the LGG occurs in every benchmark-model combination and accounts for nearly half of globally insufficient responses overall. However, conventional faithfulness verifiers that check claims against the evidence largely miss these failures: at thresholds retaining at least 95% of reliable traces, recall for LGG cases is substantially lower than that for locally unsound traces. To address these failures, we formalize three dependencies for reliable reasoning: evidence-to-step support, question-to-trace alignment, and trace-to-answer closure. Instead of post-hoc diagnosis, we introduce E-Closure to supervise these dependencies during training, combining generation supervision on supported original and counterfactual responses with bidirectional switching constraints. Averaged over three benchmarks and three backbones, existing fine-tuning baselines improve accuracy and local soundness over the base models, but at the cost of global sufficiency. E-Closure improves both: among all fine-tuned methods, it achieves the highest average accuracy (92.8%) and trace reliability (89.0%) while yielding the lowest LGG rate (6.2%). --- ### Hearsay: Can an Auditor Trust the Record a Deployed Agent Harness Writes? Authors: Jiahong Dai, Zhuochen Yang, Pengyang Shao, Kelvin Ng, Zhongyi Liu, Chengquan Ju, Yuting He, Bo Hu Published: 2026-09-26 Categories: cs.CR, cs.AI, cs.CE, cs.SE arXiv: https://arxiv.org/abs/2609.32495v1 An agent harness, the code that turns a model into an agent, writes its own record of each run, and that record is all a later reader gets when a run is disputed, investigated or audited. We call a record evidentiary when a reader who was not there can check it without trusting the writer. Across sixteen deployed frameworks, none writes one in full. Hearsay examines the record, not the task: five harnesses run fourteen tasks, three blinded LLM examiners and a human panel read the records, and every excerpt an examiner quotes is checked mechanically for who wrote it. First, the record lets a reader name the fault but not prove how the run went. Examiners name the right fault in 74 to 91% of 140 runs, but the fault can be proved only from two files the benchmark adds; for what happened in between, fewer than one citation in ten lands on anything the harness did not write, and the examiner with the fewest false alarms catches half of the entries we delete, rewrite or fabricate. Second, the remedy is a second author, not a stronger seal on the first. An append-only log of what passes between harness and model, kept outside the harness and read against the record in both directions, reports all 28 omissions and fabrications we made a harness commit as it ran, where a hash chain over the harness's own record passes all 28. Handed the log, examiners keep their fault verdicts but rest more of their citations on what the harness did not write. What makes a record evidence is who writes it, not what is captured. --- ### SoFT: Soft Targets for Generalizable LLM Fine-Tuning Authors: Huihao Jing, Wenbin Hu, Shaojin Chen, Haochen Shi, Zhongwei Xie, Guijia Zhang, Yuxuan Liu, Haoyu Huang, Haoran Li, Yangqiu Song Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32493v1 Distillation enables student language models to acquire new capabilities from expert teachers. However, integrating knowledge from multi-teacher, multi-domain demonstrations into a single student remains challenging. We study supervised fine-tuning (SFT) in this setting, where students must acquire diverse capabilities while maintaining generalization beyond the training tasks. Our experiments reveal varying trade-offs between in-distribution learning and out-of-distribution generalization across SFT methods, motivating more explicit control over this balance. To this end, we propose soft-target fine-tuning (SoFT) to balance learning from teacher demonstrations with retaining the Base model's existing capabilities. SoFT sets a minimum target probability for each demonstrated token while making the smallest KL change to the Base distribution. The resulting objective couples learning from demonstrations with adaptively weighted regularization toward the Base model. We further use domain-specific gradient budgets to control this balance and determine a probability threshold for each trajectory. Experiments on mixed-domain reasoning and agentic tasks show that SoFT achieves the best overall performance among the compared methods, with improvements in both in-distribution capability acquisition and out-of-distribution generalization. --- ### Beyond Prompt or Skill? Attribution-Guided Optimization of Modular LLM Programs Authors: Haoran Shou, Haoyue Liu, Yu Huo, Kun Zeng, Xiaoying Tang Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32492v1 Large language models can solve increasingly diverse reasoning tasks, yet their performance remains highly sensitive to task prompts, intermediate instructions, and the way reusable problem-solving knowledge is incorporated. Existing optimization methods usually focus on only one part of this design space: they either optimize a monolithic prompt, or separately induce and refine skills from model traces. As a result, they lack a principled mechanism for deciding which component should be updated when failures occur, and they rarely optimize prompts, skills, and skill-use policies in a unified framework. We propose SPARO (Skill, Prompt, And Routing Optimization), a framework that jointly optimizes task instructions, reusable skill blocks, and routing rules. It performs controlled counterfactual evaluations, converts examples' effects into a probabilistic responsibility distribution over prompt, skill, and routing components, samples one component from that distribution, and applies the corresponding targeted mutation. This design moves language-program optimization beyond global prompt rewriting toward structured, reusable, and selectively activated task knowledge. Across five benchmarks and five worker models, SPARO consistently outperforms both prompt-centered and skill-centered optimization baselines. These results suggest that effective language-program optimization depends not only on discovering useful task knowledge, but also on deciding where that knowledge should be stored and when it should be activated. --- ### Explaining Textual Entailment with Lexical Entailments: Using LLMs to Supply Lexical Relations for Formal Proofs Authors: Jorryt de Jong, Stefan Moraca, Ettore Cesari, Lasha Abzianidze Published: 2026-09-26 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.32491v1 Large Language Models (LLMs) are highly capable of natural language reasoning and appear to store a great deal of lexical knowledge, but it is still unclear how much of this knowledge they actually use when reasoning, and whether they use it in the right way. On the other hand, logic-based Natural Language Inference (NLI) systems provide transparent and formally grounded reasoning, but they need to be supplied with rich lexical knowledge to prove inferences beyond purely logical ones. In this paper, we evaluate whether LLMs can identify all lexical knowledge needed to solve NLI problems and how much this knowledge contributes to proof search in a logic-based NLI system. Our research focuses exclusively on structured lexical entailments (e.g., chinchilla$\sqsubseteq$small animal) as a proxy for structured explanations for NLI problems with an entailment label. First, we curate a dataset for a new task of explaining sentential entailments with a set of lexical entailments. The dataset is used to intrinsically evaluate LLMs on generating structured lexical explanations. Then, we use NLI as an extrinsic evaluation in a simple neuro-symbolic setting, assessing whether LLMs can supply sufficient lexical relations to LangPro, a natural-logic theorem prover for natural language. The results show that the proposed task remains challenging even for hosted proprietary LLMs, and that their contribution to theorem proving is moderate: generated relations are often only partially sound and may be tailored to the specific NLI problem rather than representing generally valid lexical knowledge. --- ### RepoMAS: Solving Progressively Specified Tasks with Issue-Driven Multi-Agent Systems Authors: Yuchen Song, Andong Chen, Wenxin Zhu, Muyun Yang, Tiejun Zhao Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32490v2 LLM-based multi-agent systems (MASs) have shown strong potential for solving complex tasks, but most assume that task requirements are sufficiently specified before execution. In practice, user requests are often incomplete, and additional requirements may only become clear during reasoning, tool use, or execution. We refer to such problems as progressively specified tasks. To systematically study this setting, we introduce ProgSpec, a benchmark that evaluates final outputs against requirements explicitly stated in the initial request and additional requirements supported by the available task evidence. We further propose RepoMAS, an issue-driven multi-agent framework inspired by open-source project management. RepoMAS records newly discovered requirements, conflicts, and failures as structured Issues and uses them to revise the task specification and execution structure during problem solving. Across ProgSpec and five existing benchmarks, RepoMAS achieves the best performance. Further analyses show that its issue-driven revision and repository maintenance mechanisms consistently contribute to performance. These results highlight the importance of allowing MASs to revise not only how a task is solved, but also revise their explicit representation of task requirements during execution. --- ### When Helpful Text Hurts: Option-Redirecting Bias in Vision-Language Models Authors: Tam Le Thi Thanh, Hoang Tran Van, Hong-Hanh Nguyen-Le, Thanh Duc Ngo Published: 2026-09-26 Categories: cs.AI, cs.CV arXiv: https://arxiv.org/abs/2609.32489v1 In tri-modal visual question answering (VQA), auxiliary text is commonly used to complement visual and textual inputs, yet its reliability is often uncontrolled. While prior work studies modality conflicts in general, it remains unclear how different types of unreliable auxiliary text affect answer selection under fixed image-question-option contexts. In this work, we show that the most harmful auxiliary text is not necessarily the most factually incorrect, but the one that aligns with the question while contradicting the image and favoring a specific distractor, leading to systematic redirection of model predictions. To isolate this effect, we introduce the Textual Reliability Ladder, a controlled diagnostic protocol that decomposes auxiliary text along three axes: image consistency, question relevance, and option support. Across multiple datasets (ScienceQA, VCR, A-OKVQA, Causal-VidQA) and recent VLMs, we find that such distractor-supporting text induces the largest accuracy drops (up to 53.1%) and concentrates errors on specific incorrect options. To mitigate this failure mode, we propose a training-free inference-time intervention that explicitly counteracts this redirection effect via noise-stability steering and dynamic grounding, reducing redirected errors while largely preserving performance under faithful text. Our results highlight that auxiliary-text reliability must be understood at the decision level, rather than solely through factual correctness, and provide a practical pathway toward more robust tri-modal reasoning. --- ### When Does Dense Retrieval Need Asymmetric Geometry? A Bias-Variance Theory of Shared and Dual Projections Authors: Maojun Sun, Yancheng Yuan, Jian Huang, Ruijian Han Published: 2026-09-26 Categories: cs.IR, cs.LG arXiv: https://arxiv.org/abs/2609.32488v1 Dense retrieval powers retrieval-augmented generation, semantic search, and question answering, yet the theoretical basis for choosing between shared and dual query-document projections remains unclear. We introduce a bias-variance theory for low-rank bilinear scoring. Shared projections induce positive-semidefinite operators, whereas dual projections realize arbitrary low-rank operators. We derive their exact approximation gap and prove a local Gaussian boundary: dual has lower risk exactly when squared directional signal exceeds the estimation cost of its additional degrees of freedom. This boundary motivates the Cross-fitted Asymmetry Risk Selector (CARS), which estimates reproducible directional signal from training pairs; its Gaussian counterpart admits exact selection-power and regret formulas. Guided by the theory, we run retrieval experiments across multiple datasets and embedding models. The mean Dual-minus-Shared NDCG@10 advantage more than doubles as query rotation increases from 0 degrees to 90 degrees. In the rank-sample-size grids, Shared wins 13 of 16 cells at n=32, whereas Dual wins all 32 cells at n=1024 and n=2048. Consistent with this shift, all 168 comparable operator-risk curves move toward Dual as training data grow. Compared to the two fixed-geometry baselines, CARS reduces held-out regret by 49-96% and achieves 90.1% mean geometry-selection accuracy. --- ### Neural Dynamics as the Composition of Quantized Units Authors: Jacopo Minniti, Aravinth Kulanthaivelu, Richard Sproat Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32487v1 Deep learning is commonly interpreted at two levels: the macroscopic, through aggregate trends in loss summarized by scaling laws, and the microscopic, through neurons, features, and circuits. A central challenge is understanding how these levels connect, so that we can explain how elementary computations compose and collectively shape macroscopic behavior. To this end, we study an intermediate abstraction in which training is described as the ordered acquisition of quanta: reusable computations acquired suddenly and binary-activated across examples to reduce loss. By approximating population-gradient updates, we derive quanta's acquisition dynamics. This yields an acquisition priority governed by demand, how frequently a computation is required across examples, and conditional complexity, how difficult that computation is to acquire given those already available. In a Boolean compositional task, we derive predictions for acquisition order and show how staggered discrete acquisitions can produce smooth aggregate loss and, under certain geometries of quanta composition, give rise to scaling laws. We then train a Transformer to map numerals to English number names and recover candidate quanta from its checkpoint trajectory. From these units, we construct a model that preserves much of the Transformer's behavior while exposing interpretable latent computations and acquisition dynamics consistent with the theory. Separately, the quanta structure can serve as training targets to improve transformer generalization. Together, these results suggest the quanta abstraction can provide useful computational atoms for studying a variety of macroscopic phenomena. --- ### Elastic Selective Spectral Hybrids for Train-Once, Export-Many Budgeted Inference Authors: Dachuan Song, Chuchu Chen, Xuan Wang Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32486v1 Deploying a language model under different computing and latency budgets calls for compact models with different quality-cost trade-offs. Towards this end, elastic spectral state space models provide ordered and temporally decomposed channels that can be truncated, but they are associated with linear time-invariant filters that cannot selectively preserve relevant past information or forget irrelevant information as the context evolves. To address this, we introduce the Elastic Selective Spectral Hybrid (ESSH), which realizes each Hankel spectral channel as an independent recurrent unit using fitted damped rotation modes. It also features an input-dependent decay and write/read gates that make temporal retention and state update input-dependent while preserving channel-wise truncation and a structured recurrence for efficient execution. ESSH combines these selective spectral mixers with sliding-window attention and jointly trains multiple capacities by reducing spectral-channel count and feed-forward width at different rates through a two-rate capacity map with full-model distillation. The resulting models support chunked parallel training and fused recurrent decoding while avoiding computation for discarded channels. At full capacity, ESSH achieves language-modeling quality comparable to similarly sized independently trained models, while smaller exports exhibit a smooth quality-cost trade-off. We validate the effectiveness of the proposed framework using language understanding, retrieval, cross-domain text, and DNA experiments, by assessing quality retention and the trade-off against independently trained and elastic baselines. At the 1.53B model configuration, fused batch-one decoding takes 1.37 ms per token on a B300, providing a 2.14-2.80x speedup over the tested Mamba-2 and Mamba-3 implementations and 3.03x over Transformer++ at matched parameter counts. --- ### What Should Federated LoRA Share? FedSAIL via Input-aware Subspace Alignment Authors: Junye Du, Shuaida He, Long Feng Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32485v1 Federated low-rank adaptation (LoRA) requires identifying an update structure that is shared across heterogeneous clients. Prior work reports strong similarity among trained LoRA projection matrices across clients; however, such agreement may be largely induced by common initialization and collapses toward random overlap under independent initialization. More crucially, relying solely on parameter similarity inherently ignores the influence of local input regime. To uncover a more robust shared structure, we introduce an input-aware action matrix that weights the adapter update by the second-moment statistics of local layer inputs. Empirically, while parameter similarity vanishes, the leading right singular directions of this action matrix remain strongly aligned across clients. This shared geometry preserves task-conditioned differences and naturally varies across network depths. Motivated by these findings, we propose Federated Subspace-Guided Action-Informed Learning (FedSAIL). Instead of averaging weights, FedSAIL estimates a shared action subspace to regularize local training while preserving client-specific coefficients. Across several benchmarks, our approach consistently improves predictive performance over competing federated LoRA methods while reducing communication cost significantly. --- ### Towards Scalable Data Diversification for Language Model Pretraining via Leverage Score Sampling Authors: Zailin Ma, Quzhe Huang, Yujun Li, Congyuan Rao, Yaodong Yang Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32484v1 Data selection for language model pretraining faces a fundamental tension between quality and diversity. While quality filtering is empirically effective, it often induces diversity collapse: by favoring texts similar to high-quality reference corpora (e.g., educational or QA-style data), it systematically excludes valuable data from underrepresented domains. In contrast, diversified selection preserves domain balance and encourages robust downstream performance, yet existing methods either focus on coverage-oriented objectives that indirectly enhance diversity, or directly optimize for diversity via costly covariance matrix recomputation that limits scalability. To address these issues, we introduce \textbf{Leverage Score Sampling (Lev)}, which iteratively selects samples that maximally expand the determinantal volume of the embedded data via leverage scores, a computationally efficient criterion that eliminates matrix recomputation and enables scalable selection. Empirically, Lev delivers up to $72\times$ speedup and improves dataset diversity, measured by the Vendi score, by $9.2\%$ over the strong diversification baseline \textbf{DiSF}. On CommonCrawl (CC) web data selection, Lev improves accuracy across seven downstream tasks by up to $1.31\%$ over existing baselines. For domains where robust quality criteria are inherently difficult to define (e.g., code), Lev serves as an effective unsupervised curation alternative: on StarCoderData, the selected subset reduces bits-per-byte by $3.08\%$ over DiSF. Notably, we uncover a cross-domain collapse of quality filtering: CC data filtered by DCLM-fastText fail to retain sufficient code-related content, yielding inferior code performance relative to Lev-selected data. These findings advocate for integrating diversity-aware practices into quality filtering for more effective data curation in language model pretraining. --- ### Separating Diagnosis from Disease Representation: Dual-View EEG Learning with Neural-Dynamics-Guided Deformation Authors: Jiaying Wang, Shouqian Shi, Yutong Chen, Xu Yang, Jie Chen, Xingyu Pan, Lei Zhang, Sheng Zhong Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32483v1 Electroencephalography (EEG)-based closed-loop neuromodulation calls for a subject-specific structured state, as opposed to a single disease probability, specifying which brain regions are deviant, at which frequencies, and at which lags. Sensor-space models keep the strongest diagnostic evidence without anatomy, source-space models give anatomy at a loss of predictive signal, and post-hoc attributions stay outside the prediction. We separate the two instead of forcing them into one representation, and propose DMD-EEG (Dual-view Multiscale Deformation for EEG), which keeps a fixed scalp spectral expert for diagnosis and models the source-space disease-related representation as a low-rank, sparse, iterative deformation of a healthy neural-dynamics prior in a $46$-region-of-interest (ROI) $\times$ $5$-frequency $\times$ $4$-lag (autocorrelation-timescale) space. The two experts meet only at a fixed decision level, so the source state is architecturally separate from the scalp expert. Across major depressive disorder (MDD), first-episode psychosis (FEP), and Parkinson's disease (PD), decision-level fusion matches the strongest single expert on MDD and FEP and exceeds the source branch on PD. On FEP the source expert is the strongest branch, the task where the deformation contributes most. The source state is an explicit ROI-frequency-lag attribution defined in a shared source coordinate system across montages, which we treat as an anatomically-coordinated predictive representation whose coordinates are directly readable and hypothesis-generating. The highest-saliency coordinates align with established disease circuitry (fronto-limbic-temporal regions in MDD, motor-cortex beta in PD), and the MDD state transfers by rank to an unseen cohort recorded with a different montage. --- ### From Outcomes to Strategies: Learning Strategy Utility for Mathematical Reasoning Authors: Ruikang Zhang, Xiao An, Xuli Shen, Jiaxing Sun, Xiaoyi Yu, Jin Zeng, Jiang Wu, Tong Lin Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32482v1 Reinforcement learning with verifiable rewards has substantially improved mathematical reasoning. However, terminal correctness alone provides limited insight into the quality of high-level strategies, such as theorem selection and subgoal decomposition, when considered separately from their subsequent execution. This paper studies strategy utility, which is defined as the likelihood that a strategy supports a correct downstream solution under a given executor. We introduce SURE, a framework for learning and leveraging relative strategy utility. In this framework, high-level strategies are separated from their detailed reasoning. Based on the pairwise preferences constructed from strategy-conditioned rollouts and teacher-generated contrasts, a Strategy Reward Model is learned to estimate relative strategy utility. During reinforcement learning, the frozen reward model reads only the extracted strategy, whose score is combined with the correctness and format rewards in a sequence-level GRPO objective. Compared with outcome-and-format GRPO baselines, experiments show that SURE improves average pass@1 by 1.87%, 2.64%, and 2.93% across three policy backbones. Our method also achieves competitive or better accuracy than stronger reward baselines while requiring substantially lower GRPO-stage compute. --- ### JEPA Learns What the Mask Leaves Unrecoverable Authors: Peng Xie, Amr Alanwar Published: 2026-09-26 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2609.32481v1 Joint-embedding predictive architectures are unusually sensitive to how the input is masked: block masks work, scattered masks do not, and the explanations are empirical. We give a measurement account. A mask is a linear measurement, and in a compactly supported wavelet basis every atom whose support lies inside the hidden region falls in the measurement's null space and leaves no trace in the data. The JEPA loss asks only that the encoded context suffice for the target, so a target that a low-level prior can recover admits a shortcut, one the moving-average target encoder can make self-consistent. What removes the shortcut is the coarse-scale content the mask leaves unrecoverable, provided enough context stays within reach of each target. We score that content before training and test the account's distinctive predictions in 151 pre-training runs. On ImageNet-100, strip masks match blocks in area and contiguity yet are recoverable, and they land at 40.3% linear top-1, beside random masks at 40.8%, against 64.3% for blocks; within one geometry family, the placements that leave the least unrecoverable content lose 6.5 points to those that leave the most, over five seed pairs; pixel targets span 7 points where latent targets span 25; and against a frozen target the gap between random and block masks, 19 points on the same kind of GPU, closes to 1.5, so the geometry acts through the target the encoder produces for itself. On UCF101 the masking ratio decides which condition, content or reach, binds; removing whole frames, unrecoverable in space but recoverable from neighbouring frames, is worst at both ratios; and on V-JEPA's own masks, batching them intact instead of truncated changes little (36.0% against 35.1%), whereas making 100 target tokens inside the blocks visible lifts them to 48.7% and hiding 100 context tokens outside the blocks does not (33.7%). --- ### Fast and Precise Learned Charged-Particle Trajectory Regression at the Large Hadron Collider Authors: Jonathan Renusch, Benjamin Huth, Daniel Murnane, Eleni Xochelli, Doğa Elitez, Paul Gessinger-Befurt, Andreas Stefl, Jeremy Couthures, Andreas Salzburger, Lukas Heinrich, Michael Kagan, Markus Elsing Published: 2026-09-26 Categories: cs.LG, hep-ex, hep-ph, physics.data-an arXiv: https://arxiv.org/abs/2609.32479v1 We propose a training recipe that treats charged-particle trajectory parameter regression on high-energy physics detector data as a sequence-modeling task. Kalman filters and linearized least-squares fits have been the classical standard approach for this task: they are optimal estimators for sparsely sampled linear-Gaussian data and are commonly used for trajectory parameter regression (fitting). The classical fitting techniques implemented for this domain reach a final precision of one part in $10^5$ through detailed modeling of detector geometry, material, detection effects and precise numerical integration of the equations of motion through the detector's inhomogeneous magnetic field. With this study, we demonstrate that using a bidirectional gated linear recurrent encoder, one is able to reproduce the full precision of classical track fitting techniques. Using a custom kernel, we also achieve significantly higher throughput during GPU inference, compared to classical fitting software running on similarly priced multi-core CPU servers representing typically employed hardware. Such a speedup would lead to considerable cost savings for the pattern recognition at the Large Hadron Collider. To our knowledge, this is the first end-to-end learned track fit to reach the full precision and, at the same time, offer the opportunity to reduce the computing costs. --- ### Agnostic Smoothed Online Regression with Adversarial Responses Authors: Xuanyu Chen, Yue Yu Published: 2026-09-26 Categories: stat.ML, cs.LG, math.ST arXiv: https://arxiv.org/abs/2609.32478v1 We study smoothed online prediction with bounded adversarial responses. This widely studied framework bridges i.i.d. sampling and adversarial covariate selection through a smoothness parameter $\textsf{C}_{\textsf{cov}}$, which bounds conditional covariate densities relative to a fixed, unknown base measure. We propose \textsc{Hedge-Cover}, an information-theoretic algorithm that achieves sublinear regret $\widetilde{O}(\sqrt{\text{Pdim}(\mathcal{F}) \textsf{C}_{\textsf{cov}} T})$ for function classes with bounded pseudo-dimension. The algorithm aggregates a carefully constructed family of experts using \textsc{Hedge}, with a prior that links regret to the number of disagreements between a consistent selector and a target function. We bound this number by exploiting covariate smoothness. This answers an open problem posed in \cite{blanchard2025agnostic} on the minimax optimal adaptive regret of the smoothed online regression problem. We establish a matching lower bound for the class of linear predictors. The main intricacy of the lower bound lies in explicitly constructing a challenging sequential covariate distribution supported on mutually orthogonal hyperplanes. This construction may be of independent technical interest. Finally, we revisit the well-specified setting and quantify the effect of response noise. For conditionally $ν^2$-subGaussian responses, we extend the existing lower bound under realizable responses by showing that the minimax expected regret is $Ω((1\vee ν)\sqrt{(\textsf{C}_{\textsf{cov}}-1)dT})$ for a function class of VC dimension $d$. A corresponding upper bound for ERM matches this dependence on $ν$. --- ### Back-Tracking from Clarity: Self-Learning to See Text from Afar Authors: Duc-Tri Tran, Phi Le Nguyen, Minh Hoai Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32477v1 We propose a self-supervised framework designed to enhance the capability of scene text detectors in identifying and recognizing text in scenarios where instances are shown at significant distances, typically small, blurred, and frequently missed by conventional models. Our approach leverages the high-fidelity performance of existing text spotting models on large, clear text as a foundational supervisor. By temporally back-tracking these high-confidence detections through video sequences, we automatically synthesize pseudo-labels for preceding frames where the distant text is still visually degraded or undersized. These pseudo-labels enable training a student model specialized for early text detection, without requiring any manual annotation. The success of this approach depends on accurate pseudo-label generation, for which we develop a dedicated scene text tracker capable of maintaining consistent text identities across challenging video sequences. In addition, we propose SceneText50, a diverse multilingual outdoor dataset to facilitate training and evaluation. Experiments show that our framework significantly improves early detection accuracy and robustness across varied scenes and languages. Code and data are at \href{https://github.com/trid2912/BackTrackingText}{https://github.com/trid2912/BackTrackingText}. --- ### PC-SubMax: Efficient Prompt Compression via Regularized Submodular Maximization Authors: Ziyi Zhang, Shuang Cui, Haotian Zhang, Xiaoyu Wang Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32474v1 While large language models (LLMs) are increasingly deployed in long-context scenarios, lengthy prompts can increase inference costs and latency and exacerbate the ``lost-in-the-middle'' phenomenon. Selective prompt compression offers a model-agnostic approach to alleviating these issues. However, methods based on fixed token- or sentence-level importance scores may overlook how content contributions change with the selected subset, limiting their ability to account for inter-sentence redundancy. Compression procedures that rely on autoregressive LLM scoring can also introduce substantial overhead. We propose PC-SubMax, a theoretically grounded framework that formulates selective prompt compression as regularized monotone submodular maximization under a knapsack constraint. The objective is $U(S)-\ell(S)$, where the monotone submodular utility $U$ combines information coverage, query relevance, and log-determinant diversity, and the non-negative modular penalty $\ell$ captures token cost. Through diminishing marginal returns, the objective evaluates each sentence's contribution relative to the selected content. To optimize this objective, we develop the Regularized Greedy+Max (RGM) algorithm, which deterministically returns a feasible set $Q$ satisfying $U(Q)-\ell(Q)\geq \frac{1}{2}U(O)-\ell(O)$, where $O$ is an optimal feasible solution to the regularized problem. RGM uses $O(nκ)$ value-oracle queries, where $n$ is the number of candidate sentences and $κ$ is the maximum feasible subset size. PC-SubMax uses encoder representations and avoids autoregressive LLM scoring during compression. Experiments across seven diverse benchmarks demonstrate competitive downstream performance with low compression overhead. --- ### VPEvolve: A Self-Evolving Virtual Process Engineer for Computational Lithography Authors: Tianyi Li, Wenxuan Dong, Donger Luo, Nan Wang, Yanpeng Chen, Jiaqi Liu, Xinyun Zhang, Hao Geng Published: 2026-09-26 Categories: cs.AI, cs.SE arXiv: https://arxiv.org/abs/2609.32473v1 Optical proximity correction (OPC) recipes grow as engineers add local rules to repair newly discovered lithography hotspots. Each correction can interact with existing rules, while lessons from commercial-tool trials remain scattered across code and logs. \system combines a Virtual Process Engineer (VPE) harness with a Skill Bank of measured engineering experience. The harness equips a frozen language model with process manuals, layout analysis, recipe editing, and commercial-tool evaluation. The actor proposes changes to the global parameters, local targeted rules, or diagnostic trials. After each evaluation, an LLM reflector and curator turn the measured response into evidence-linked judgments. The actor retrieves them before its next trial. Feasible improvements update the retained recipe; every measured trial informs the Skill Bank that guides the next edit. The model weights remain fixed. On a FreePDK45-derived benchmark with ten commercial-tool evaluations per case, \system reduces the mean per-case maximum edge placement error from 18.294 to 5.361 nm on Poly and from 22.052 to 15.692 nm on Metal1. Every final recipe satisfies the predefined quality constraints and improves the maximum error by at least 0.1 nm. --- ### AdaTutoRank: Learning to Rerank Document Sets via Adaptive Tutoring Optimization for RAG and Deep Research Authors: Kailin Jiang, Lei Liu, Jian Xi, Yangqi Chen, Hui Xu, Hongwei Zhao, Bin Li, Yu Lu, Haibo Shi Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32472v2 Document rerankers determine what evidence reaches the downstream model in RAG and deep research, yet mainstream rerankers select by relevance matching, and individually relevant documents rarely constitute the complete, complementary, non-redundant set a complex information need demands. Prior work rewards a set by its aggregate rubric score, shifting the objective from ranking documents to composing sets. Yet that score is one scalar shared by every document in the set, so the supervision is sparse: a redundant document is rewarded with the rest whenever the set scores well, and a decisive one penalized with the rest whenever it does not; credit assignment leaves contributors indistinguishable from free riders. On-policy distillation could densify this supervision, but existing methods give every rollout the same fixed guidance, too prescriptive for strong rollouts and too abstract for weak ones. We therefore propose AdaTutoRank, a setwise reranker trained with Adaptive Tutoring Optimization (ATO) under a three-level hierarchy of nine rubric dimensions, which supplies silver labels for the cold start, rewards for reinforcement learning, and hints for distillation. ATO draws three hint forms of increasing specificity from the policy's own frozen snapshot: the rubrics alone, a self-selector's sibling-set chosen under rubrics, and a self-reflector's reflection contrasting the rollout with that sibling-set; each rollout receives the form matched to its quality. Re-scoring that rollout under the hint-conditioned frozen teacher and the hint-free snapshot distills the hint's effect into a token-level advantage that complements the group-relative outcome advantage. Across ten benchmarks spanning RAG, deep research, and setwise evaluation, AdaTutoRank attains the best overall performance while issuing fewer retrieval calls. --- ### On the Pitfalls of Verbalized Confidence Priors for Calibrating Large Reasoning Models Authors: Shuoyuan Wang, Beier Luo, Hao Zeng, Chengyao Yu, Songxin Zhang, Zejian Xie, Bingyi Jing, Hongxin Wei Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32470v1 Large reasoning models (LRMs) often suffer from overconfidence when expressing their uncertainty. Confidence-aware reinforcement learning (RL) offers a promising way to optimize calibration. However, it relies on on-policy rollouts and is thus constrained by the model's pre-RL confidence distribution, which we term confidence prior. In this work, we reveal that off-the-shelf LRMs exhibit a confidence prior heavily concentrated on a few high values, which persists throughout RL. Theoretically, we prove that this concentration suppresses policy gradient updates for rarely sampled confidence values and inflates the lower bound on expected Brier risk. To overcome this exploration bottleneck, we propose CalibSFT, a plug-and-play supervised fine-tuning stage that shapes a calibrated confidence prior with broad support before RL. For each question, CalibSFT constructs confidence targets combining its success rate with response-level correctness, which provably preserves proper-scoring optimality, and then balances training responses across the confidence spectrum to enable diverse confidence exploration during RL. To learn from incorrect responses without imitating their reasoning, CalibSFT introduces correctness-conditional supervision, guiding confidence across all responses while supervising reasoning only on correct ones. Across 16 mathematical and general reasoning benchmarks, incorporating CalibSFT reduces calibration errors and improves discrimination across five representative RL algorithms while preserving comparable accuracy. Furthermore, CalibSFT delivers practical benefits for downstream selective prediction and model routing. Our code is available at https://github.com/ml-stat-Sustech/verbalized-confidence-training. --- ### PULSE: Identifying Demonstration-Utility Features with Sparse Autoencoders Authors: Chenduo Hao, Chuanbao Gao, Pinjun Zeng, Jingze Zhu, Chonghan Liu, Zidong Liu, Xu Yang Published: 2026-09-26 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.32469v1 In-context learning is highly sensitive to demonstration choice, yet most methods select demonstrations using external query-demonstration similarity. Such criteria can miss model-specific signals: Similar demonstrations may activate different internal features and downstream behaviors. We introduce PULSE (Paired Utility Localization over Sparse Encodings), an SAE-based framework for identifying model-internal features associated with demonstration utility and using them for demonstration selection. Using a small labeled discovery set, PULSE samples candidate demonstration sets, measures their zero-shot-relative utility under the target model, and scores SAE features by how their activation differences align with utility differences. The top positive and negative coordinates form a sparse utility-localization vector. We use this vector in two complementary ways: as a signed score for controlled complete-set ranking, and as PULSE-Retriever, which converts its magnitude into a feature-relevance mask for scalable pool-scale retrieval. Across classification, generation, and reasoning benchmarks, PULSE-Retriever improves over the strongest baseline by 2-3 accuracy points, 0.6-0.9 BLEU-4, and 3.2 exact-match points, respectively, while controlled ranking validates the identified features encode a predictive set-level utility signal. Feature inspection and cross-dataset experiments suggest that the identified features capture task-relevant, dataset-conditioned patterns, yet retain utility signals that partially transfer across datasets. Our code is available at https://github.com/aohenuo/PULSE. --- ### Bison: Cross-Dataset Learning for Unseen-Compound Perturbation Prediction Authors: Yunfan Liu, Kasra Ghorbani, Yufei Huang, Zicheng Liu, Jiangbin Zheng, Jingbo Zhou, Shaorong Chen, Chang Yu, Stan Z. Li Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32467v1 Predicting transcriptional responses to unseen compounds is limited by fragmented chemical coverage and heterogeneous experimental platforms and gene panels. To assess molecular generalization across these settings, we build on Chem-PerturBridge to benchmark eight datasets with 16,771 compounds, withholding test compounds from every training dataset. This comparison reveals that high overall response agreement can coexist with weak prediction of drug-specific differences, despite reproducible signals across repeated measurements. To exploit complementary chemical supervision while targeting these differences, we introduce Bison: a shared gene representation connects native panels, while two discrete diffusion models compose context-dependent responses with molecular deviations learned through matched drug-contrast supervision. A single Bison model jointly trained across all eight datasets achieves the highest mean overall-response and drug-contrast Pearson correlations on the full benchmark in comparison with 11 methods trained independently per dataset. Compared with dataset-specific training of the same architecture, joint training increases mean drug-contrast correlation by 27.4\%, with gains across all eight datasets and improvements in overall response prediction. These results demonstrate how matched drug contrasts turn complementary screens into shared molecular supervision for unseen-drug response prediction while preserving native gene measurements. --- ### PolyStepOR: Learning to Decide Without Optimal Decisions Authors: Viet The Nguyen, Gunther Gust, An Thai Le Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32465v1 Decision-focused learning (DFL) trains predictors for downstream decision quality, but often relies on optimal reference decisions that are expensive to obtain. We present PolyStepOR, which trains directly from realized decision costs without pre-computed optima and extends to in-constraint predictions through repair or infeasibility penalties. To handle piecewise-constant losses, PolyStepOR perturbs predictor parameters, evaluates the resulting decisions, and uses optimal transport to favor lower-cost directions, requiring no derivatives. Without task-specific tuning, PolyStepOR performs strongly on classical optimization benchmarks and competitively on predicted-constraint and real-world problems. Theoretically, we characterize decision-preserving perturbations and boundary detection, bound sensitivity to cost errors, and establish stationarity guarantees for a smoothed objective. PolyStepOR thus replaces optimal reference decisions and derivatives with forward evaluations. ---